Stephen BlumBig matrix values can explode gradients and stop learning fast.
kipriyanovichAlright, future engineers!
The **Determinant** of a square matrix is a scalar value that tells you if the matrix is invertible.
Ex: For `[[a,b],[c,d]]`, `det = ad - bc`.
Pro-Tip: If `det = 0`, the matrix is singular (no inverse), meaning `Ax=b` has no unique solution!
Alright, future engineers!
**Matrix multiplication** combines two matrices, forming a new one where elements are dot products of rows & columns.
Ex: `(AB)_ij = sum(A_ik * B_kj)`. Inner dimensions must match!
Pro-Tip: It's NOT commutative (AB != BA)!
#Matrices #LinearAlgebra #STEM #StudyNotes
**Matrix multiplication** combines two matrices, forming a new one where elements are dot products of rows & columns.
Ex: `(AB)_ij = sum(A_ik * B_kj)`. Inner dimensions must match!
Pro-Tip: It's NOT commutative (AB != BA)!
#Matrices #LinearAlgebra #STEM #StudyNotes
Sean Cribbs 🌻Thanks for attending! We were able to port a lot more of `glam` into our Gleam matrices library. A polished release will be right around the corner and 1.0 not far off.
See you next week, like/follow/subscribe: https://youtube.com/@seancribbs
Welcome back to another Sunday! Let's polish up the https://hex.pm/packages/matrix_gleam library for a new release. Join me at 1:30PM CDT:
Miss Kitty 🌈🌈🌈Either does the #network. #Matrices of #truth which manage disconnects, spike in use, reduced use, and adoption expansion. It would be all but impossible to shut down once going and adopted even sparsely globally. It's an engine of freedom. My head is exploding with applications. Tonight's research.
Alright, future engineers!
**Matrix Multiplication** creates a new matrix where `C_ij` is the dot product of row `i` from the 1st matrix & column `j` from the 2nd.
Ex: `A(2x3) * B(3x2)` gives `C(2x2)`.
Pro-Tip: Inner dimensions must match!
grimmigesA major #update to my #collection of #morphological #matrices worthy to infer #networks:
a bunch of matrices we generated and reanalysed for a paper by S. Renner, D. Sokoloff, and me dealing with ancestors, hard polytomies and seed plant evolution; depicted not as the usual cladograms (where ancestor-descendant pairs must trigger unsolvable hard trichotomies) but as a "Romerogram" (or spindle graph) that shows #dichotomy as well as #buddingEvolution, i.e., #phylogeny
N-gated Hacker News🎩🤓 Behold, the blog that bravely attempts to explain the mystical Q, K, V #matrices as if they're the holy trinity of #AI. Spoiler alert: it's mostly a glorified game of 'which word matters?' because we all need another reason to overcomplicate common sense. 🙄🔍
https://arpitbhayani.me/blogs/qkv-matrices/ #QKV #Overcomplication #CommonSense #HackerNews #HackerNews #ngated
https://arpitbhayani.me/blogs/qkv-matrices/ #QKV #Overcomplication #CommonSense #HackerNews #HackerNews #ngated
The Q, K, V Matrices
At the core of the attention mechanism in LLMs are three matrices: Query, Key, and Value. These matrices are how transformers actually pay attention to different parts of the input. In this write-up, we will go through the construction of these matrices from the ground up.
royalWhat are your favorite programs that specialize in working with tables? My use case: building human-readable matrices for documentation and reference. Not a database.
I don't want a typical Office program (Not Excel or Word or the FOSS equivalent.)
For example, OmniOutliner or Logseq are much better for outlining than Word etc. No, I'm not going to learn Emacs Org Mode.
Markdown tables are simple but clumsy.
I've used SQLite Browser to great effect for data., but not the right tool here.
What's actually good?